天地
In linear warmup, the first global step is **zero**,and the learing rate is **zero** too(lr=global_step/num_warmup_steps * init_lr),then nothing will be updated except momentum(m and v in Adam), especially when optimizer...
# PR type - [x] Bug Fix - [ ] New Feature - [ ] Document Updates - [ ] More Models or Datasets Support # PR information fix list...
**Describe the bug** SftArguments 数据参数如下设置，其他无关均省略 dataset=[f'{DatasetName.alpaca_zh}#100', f'{DatasetName.alpaca_en}#50', f'{DatasetName.self_cognition}#250'] custom_train_dataset_path=["./data/faq_train.jsonl"], custom_val_dataset_path=["./data/faq_valid.jsonl"], dataset_test_ratio 默认为0.01，这导致了 custom_train_dataset_path和custom_val_dataset_path传入路径时，均按照0.01进行了切分，且把custom_train_dataset_path和custom_val_dataset_path里面99%的内容作为训练，1%的内容作为验证集，不合符逻辑设定。 swift版本：commitId:845ac0ce46c4b904a809e3570e8dfb830f9b4e00
# PR type - [x] Bug Fix - [ ] New Feature - [ ] Document Updates - [ ] More Models or Datasets Support # PR information cpu环境下 无cuda，但是调用了get_max_cuda_memory方法，导致报错
在官方ant的[demo](https://lowcode-engine.cn/demo/demo-basic-antd/index.html)页面，直接拖拽一个【导航菜单】到页面，提示“menu 组件渲染异常 请查看控制台日志” 